A method for predicting the Tm value of a dissolution curve

Abstract

This invention provides a method for predicting the Tm value of a melting curve, relating to the field of biodetection technology. The method includes the following steps: constructing a Tm value labeling model and a Tm value prediction model; acquiring data information of the sample to be tested and inputting the data information into the Tm value labeling model; preprocessing the data information by the Tm value labeling model and labeling the sample to be tested with or without a Tm value; samples with Tm values ​​constitute a valid sample set; extracting several peaks from the original melting curve to obtain an original Tm value set; encoding the original Tm value set according to temperature intervals; wherein the temperature bits in the original binary vector are arranged according to temperature intervals; inputting the original melting curve from the valid sample set into the Tm value prediction model to obtain the predicted Tm value. Arranging the temperature bits according to temperature intervals allows the model to learn the correlation between adjacent temperatures, thereby more accurately capturing the influence of temperature changes on the Tm value, and thus improving the accuracy of Tm value prediction.

Description

Technical Field

[0001] This invention relates to the field of biological detection technology, specifically to a method for predicting the Tm value of a melting curve. Background Technology

[0002] Polymerase chain reaction (PCR) technology is widely used in molecular biology, medical diagnostics, and gene research. One of its core steps is melting curve analysis, which uses the melting temperature (Tm value) to assess the specificity and purity of the amplified products. However, current methods do not encode the Tm value, nor do they... have Preserving the order of Tm values ​​while discretizing temperatures does not preserve the correlation between adjacent intervals, the model cannot learn the correlation between adjacent temperatures, making it insensitive to small differences and unable to capture the impact of small temperature changes on Tm values. Traditional methods also fail to provide accurate Tm values ​​and cannot identify anomalies. There is an urgent need for a method that can provide more accurate Tm values ​​and save more manual time. Summary of the Invention

[0003] The purpose of this invention is to provide a method for predicting the Tm value of a dissolution curve, and the technical problem to be solved is how to improve the accuracy of Tm value calculation based on the dissolution curve.

[0004] This invention is achieved through the following technical solution:

[0005] A method for predicting the Tm value of a dissolution curve includes the following steps:

[0006] Based on machine learning models, construct a Tm value labeling model and a Tm value prediction model;

[0007] Obtain the data information of each sample in the sample set to be tested, and input the above data information into the Tm value labeling model; wherein, the above data information includes the original dissolution curve;

[0008] The data information is preprocessed using the Tm value labeling model described above; the presence or absence of a Tm value in the test sample is then marked using the Tm value labeling model, and the test samples with Tm values ​​constitute the valid sample set.

[0009] Several peak values ​​were extracted from the original melting curves of the test samples in the above effective sample set to obtain the original Tm value set;

[0010] The original Tm value set is encoded according to the preset temperature interval to obtain the original binary vector; wherein the temperature bits in the original binary vector are connected and arranged according to the temperature interval.

[0011] The original melting curves of the test samples from the above effective sample set are input into the Tm value prediction model to obtain the prediction binary vector; wherein, the temperature bits in the above prediction binary vector are arranged according to the interval temperature.

[0012] By comparing the original binary vector and the predicted binary vector, it can be determined whether the predicted binary vector is abnormal.

[0013] The original melting curves of the test samples in the effective sample set are divided according to a preset temperature interval, and a binary vector is generated according to the temperature position. The temperature positions are separated by a preset temperature interval, which fully considers the order of temperature. This allows the model to learn the correlation between adjacent temperatures, thereby more accurately capturing the impact of temperature changes on the Tm value and helping to improve prediction accuracy.

[0014] The Tm value labeling model preprocesses the input sample data (including the original melting curve) to mark whether the sample has a Tm value. Samples with Tm values ​​are selected to form a valid sample set, eliminating interference from samples without Tm values. This allows the subsequent prediction model to focus on the analysis of valid samples, reducing the interference of irrelevant data on the model's prediction and thus improving the accuracy of the prediction. By comparing the original binary vector and the predicted binary vector, the Tm value predicted by the Tm value prediction model is verified to be normal. If the original binary vector and the predicted binary vector are consistent, the predicted Tm value is normal; otherwise, the predicted Tm value is abnormal, and the original melting curve is marked as an abnormal curve.

[0015] Furthermore, the specific steps for constructing a Tm value labeling model based on a machine learning model include:

[0016] Obtain historical data information for each first historical sample in the first historical sample set; wherein, the aforementioned historical data information includes the original melting curve and the original label with or without Tm value label;

[0017] The historical data information of each first historical sample in the aforementioned first historical sample set is preprocessed;

[0018] The preprocessed first historical sample set is divided into a first training set and a first test set.

[0019] The historical data information from the first training set is input into the machine learning model for training to obtain the Tm value labeling model.

[0020] Input the original dissolution curve of the first historical sample in the first test set into the Tm value labeling model to obtain the test label of whether the first historical sample has a Tm value label;

[0021] The original label and test label of the first historical sample are matched. If the match is successful, the first historical sample is labeled normally; if the match fails, the first historical sample is labeled abnormally.

[0022] The machine learning model learns the mapping relationship from melting curve features to the presence or absence of Tm value labels on the training set, and then validates it on the test set to examine the learning effect and generalization ability of the Tm value labeling model. The original melting curve of the first historical sample from the first test set is input into the Tm value labeling model for testing. Then, the original labels of the first historical sample are matched with the test labels to verify the accuracy of the Tm value labeling model. This process identifies problems and shortcomings in the Tm value labeling model, providing direction for subsequent model optimization. It ensures that the final Tm value labeling model can accurately distinguish between samples with and without Tm values, filtering out effective sample data for subsequent steps, reducing the interference of invalid data on the prediction model, and thus improving the accuracy of Tm value prediction.

[0023] Furthermore, the specific steps for constructing a Tm value prediction model based on machine learning models include:

[0024] Obtain historical data information for each second historical sample in the second historical sample set; wherein, the aforementioned historical data information includes the original melting curve and the original Tm value;

[0025] The historical data information of each second historical sample in the aforementioned second historical sample set is preprocessed;

[0026] The preprocessed second historical sample set is divided into a second training set and a second test set.

[0027] The historical data information from the second training set is input into the machine learning model for training to obtain the Tm value prediction model.

[0028] The original dissolution curve of the second historical sample in the second test set is input into the Tm value prediction model to obtain the predicted Tm value of the second historical sample.

[0029] The original Tm value and the predicted Tm value of the second historical sample are matched. If the match is successful, the prediction of the second historical sample is successful; if the match fails, the prediction of the second historical sample fails.

[0030] The original melting curve of the second historical sample is input into the Tm value prediction model to obtain the predicted Tm value. Then, the original Tm value and the predicted Tm value are matched to measure the prediction performance of the Tm value prediction model on the test set, determining whether the model can accurately predict the Tm value corresponding to the new melting curve. If the match fails, it indicates that the model has failed to predict on that sample. This helps to identify problems and shortcomings in the Tm value prediction model, providing direction for subsequent model optimization and prompting researchers to adjust and improve the model to enhance its prediction accuracy.

[0031] Furthermore, the aforementioned second historical sample set consists of the first historical samples in the first historical sample set that were originally labeled with Tm values.

[0032] The Tm value labeling model and the Tm value prediction model are coordinated. In the practical application stage, the Tm value labeling model is used to identify test samples with Tm values ​​to obtain valid samples; then the Tm value prediction model is used to predict the Tm value.

[0033] Furthermore, the historical data information of each first historical sample in the aforementioned first historical sample set is preprocessed, including removing first historical samples without Ct values.

[0034] Samples without Tm values ​​indicate no viral infection; samples without Ct values ​​indicate an experimental failure, and these samples are of no reference value. The Tm value labeling model was cleaned.

[0035] Furthermore, the data information is preprocessed using the aforementioned Tm value labeling model. Specific steps include:

[0036] The aforementioned data also includes information on the data source and experimental process of the samples to be tested;

[0037] The above data is cleaned to remove information about the data source and experimental process of the sample to be tested;

[0038] The original dissolution curves in the above data were normalized.

[0039] While information about the data source and experimental process of the test samples is helpful in understanding the background and experimental conditions, it does not directly contribute to the determination of the Tm value and may even become a confounding factor. Removing this information during the data cleaning stage allows the model to focus more on the original dissolution curve data, which is closely related to the Tm value. Data in the original dissolution curve may have different dimensions and numerical ranges. Data with different dimensions can make it difficult for the model to balance the importance of various features during training, leading to slower convergence or even failure to converge. Normalizing the original dissolution curve, mapping all data to a specific interval (such as [0, 1] or [-1, 1]), eliminates the influence of dimensions, makes the features numerically comparable, and allows the model to adjust parameters more efficiently during training, accelerating convergence and improving training efficiency.

[0040] Furthermore, upon receiving the traceability signal, the data source information and experimental process information of the sample to be tested are obtained from the aforementioned data information.

[0041] For each sample, patient information, experimental parameters (well number, reagent batch, PCR instrument channel) and operation records (detection time, operator ID, original curve file) are associated. The database index enables reverse tracing of "diagnosis result → Tm value → experimental conditions" and supports rapid problem localization (such as temperature drift caused by a certain batch of reagents).

[0042] Furthermore, the preset temperature interval is 0.05℃ to 0.3℃, preferably 0.3℃, to improve the model's ability to capture minute temperature differences.

[0043] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0044] The original melting curves of the test samples in the effective sample set are divided into binary vectors of temperature bits according to a preset temperature interval. The binary vectors of adjacent temperature bits are arranged continuously, which fully considers the order of temperature. This allows the model to learn the correlation between adjacent temperatures, thereby more accurately capturing the impact of temperature changes on the Tm value and helping to improve prediction accuracy.

[0045] The Tm value labeling model preprocesses the input sample data (including the original melting curve) to label whether the sample has a Tm value. Samples with Tm values ​​are selected to form the effective sample set, eliminating the interference of samples without Tm values. This allows the subsequent prediction model to focus on the analysis of effective samples, reducing the interference of irrelevant data on the model prediction and thus improving the accuracy of the prediction. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0047] Figure 1 Main flowchart;

[0048] Figure 2 This is a schematic diagram showing the correspondence between binary vectors and temperature. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0050] First embodiment:

[0051] A method for predicting the Tm value of a dissolution curve, combined with Figure 1 This includes the following steps:

[0052] Based on a machine learning model (which can be any of the XGBoost model, RNN convolutional neural network, MLPRegressor, etc., with XGBoost model as an example below), construct a Tm value labeling model and a Tm value prediction model;

[0053] Obtain the data information of each sample in the sample set to be tested, and input the above data information into the Tm value labeling model; wherein, the above data information includes the original dissolution curve;

[0054] The data information is preprocessed using the Tm value labeling model described above; the presence or absence of a Tm value in the test sample is then marked using the Tm value labeling model, and the test samples with Tm values ​​constitute the valid sample set.

[0055] Extract several peaks (up to 5 peaks) from the original melting curves of the test samples in the above effective sample set to obtain the original Tm value set;

[0056] The original Tm value set is encoded according to a preset temperature interval to obtain a 151-bit original binary vector. The temperature bits in this original binary vector are arranged sequentially according to the temperature intervals. For example, with a temperature interval of 0.3℃, the 0th bit of the binary vector is 40.0℃, the 1st bit is 40.3℃, the 2nd bit is 40.6℃, the 3rd bit is 40.9℃, ..., the 49th bit is 54.7℃, the 50th bit is 55℃, the 51st bit is 55.3℃, the 52nd bit is 55.6℃, ..., the 150th bit is 85℃. If two peak values ​​are extracted, 55℃ and 55.3℃ respectively, then the 50th bit and the 51st bit of the binary vector are both 1 (see...). Figure 2 ).

[0057] The original melting curves of the test samples from the above effective sample set are input into the Tm value prediction model to obtain a predicted binary vector (i.e., the binary vector after encoding the predicted Tm value). Local temperature correlations are captured through a convolutional layer. The temperature bits in the predicted binary vector are arranged in a sequence of temperature intervals.

[0058] By comparing the original binary vector and the predicted binary vector, it can be determined whether the predicted binary vector is abnormal.

[0059] The original melting curves of the test samples in the effective sample set are divided into binary vectors of temperature bits according to a preset temperature interval. The binary vectors of adjacent temperature bits are arranged continuously, which fully considers the order of temperature. This allows the model to learn the correlation between adjacent temperatures, thereby more accurately capturing the impact of temperature changes on the Tm value and helping to improve prediction accuracy.

[0060] The Tm value labeling model preprocesses the input sample data (including the original melting curve) to label whether the sample has a Tm value. Samples with Tm values ​​are selected to form the effective sample set, eliminating the interference of samples without Tm values. This allows the subsequent prediction model to focus on the analysis of effective samples, reducing the interference of irrelevant data on the model prediction and thus improving the accuracy of the prediction.

[0061] One possible use case is to input the original melting curve, containing 186,138 noisy data points, into the Tm value labeling model. The Tm value labeling model predicts whether each of the 186,138 noisy data points contains a Tm value and labels it, automatically marking anomalous samples (i.e., no Tm value, multiple peaks, low signal-to-noise ratio). Then, 150,000 high-quality data points without anomalies, labeled with Tm values, are input into the Tm value prediction model, which predicts the Tm value for each data point.

[0062] The encoding process is as follows: Given a temperature bit value ∈[40,85], encoding step size s=0.3, encoding center point (Explanation: from) Start the loop until End the loop); then put Encode it into 151-bit binary, and place 1 in the distance. Recent Location, everything else is 0; calculate distance. Nearest center point encoding position The specific mathematical formula is as follows: ,in, Indicates the starting point of the encoding, here .

[0063] For example: =55.7, s=0.3, through The result of the center point encoding position closest to 55.7 is: the 52nd bit is 1, and the others are 0.

[0064] Second embodiment:

[0065] Based on the first embodiment, the Tm value labeling model was trained and tested. The specific steps for constructing the Tm value labeling model based on the XGBoost model include:

[0066] Obtain historical data information for each first historical sample in the first historical sample set; wherein, the aforementioned historical data information includes the original melting curve and the original label with or without Tm value label;

[0067] The historical data information of each first historical sample in the above first historical sample set is preprocessed. The preprocessing includes data cleaning (removing irrelevant data, i.e., data source information and experimental process information) and normalization.

[0068] The preprocessed first historical sample set is divided into a first training set and a first test set.

[0069] The historical data information from the first training set is input into the XGBoost model for training to obtain the Tm value labeling model.

[0070] Input the original dissolution curve of the first historical sample in the first test set into the Tm value labeling model to obtain the test label of whether the first historical sample has a Tm value label;

[0071] The original label and test label of the first historical sample are matched. If the match is successful, the first historical sample is labeled normally; if the match fails, the first historical sample is labeled abnormally.

[0072] The XGBoost model learns the mapping relationship from melting curve features to the presence or absence of Tm value labels on the training set, and validates it on the test set to examine the learning effect and generalization ability of the Tm value labeling model. The original melting curve of the first historical sample from the first test set is input into the Tm value labeling model for testing. Then, the original labels of the first historical sample are matched with the test labels to verify the accuracy of the Tm value labeling model. This process identifies problems and shortcomings in the Tm value labeling model, providing direction for subsequent model optimization. It ensures that the final Tm value labeling model can accurately distinguish between samples with and without Tm values, filtering out effective sample data for subsequent steps, reducing the interference of invalid data on the prediction model, and thus improving the accuracy of Tm value prediction.

[0073] Starting with acquiring historical data, the process involves preprocessing, dataset partitioning, model training, testing and evaluation, and error labeling. Each step is interconnected and influences the others. By continuously iterating through this process, model performance can be progressively optimized, ensuring that the Tm value labeling model can select effective samples. This helps improve the accuracy and stability of the entire Tm value prediction process, providing high-quality input data for subsequent Tm value prediction models. High-quality input data means that subsequent prediction models can learn based on more relevant and accurate information, thereby improving the accuracy of Tm value prediction. Through the synergistic effect of these steps, the entire Tm value prediction process is optimized at each stage, ultimately improving the accuracy of Tm value prediction.

[0074] One possible use case is to use 158,672 data points out of 186,138 as training data and 27,466 as test data. Each data point includes 151 melting peak curves, a label indicating whether it contains a Tm value, and data source traceability information. In binary classification tasks, XGBoost offers the following advantages over other models (such as logistic regression, random forest, and support vector machines): high efficiency and scalability, regularization to prevent overfitting, strong flexibility and parameter tuning, and optimization in minimizing structural risk.

[0075] The tests revealed that setting the XGBoost model's hyperparameters as follows: a maximum of 5000 trees in the ensemble, a maximum depth of 10 per tree, a learning step size of 0.05 for weight updates in each iteration, and an objective function that optimizes probability values ​​between 0 and 1, along with using a logarithmic loss function to evaluate the model's performance, yielded good results. The test results were as follows: out of 27466 data points in the test set, 3 data points containing a Tm value were predicted as not containing a Tm value, and 30 data points without a Tm value were predicted as containing a Tm value. The overall prediction accuracy on the test set was 99.88%.

[0076] A specific implementation example demonstrates the training and testing of the Tm value prediction model. The specific steps for constructing a Tm value prediction model based on the XGBoost model include:

[0077] Obtain historical data information for each second historical sample in the second historical sample set; wherein, the aforementioned historical data information includes the original melting curve and the original Tm value;

[0078] The historical data information of each second historical sample in the above second historical sample set is preprocessed. The preprocessing includes data cleaning (removing irrelevant data, i.e., data source information and experimental process information) and normalization.

[0079] The preprocessed second historical sample set is divided into a second training set and a second test set.

[0080] The historical data information from the second training set mentioned above is input into the XGBoost model for training to obtain the Tm value prediction model.

[0081] The original dissolution curve of the second historical sample in the second test set is input into the Tm value prediction model to obtain the predicted Tm value of the second historical sample.

[0082] The original Tm value and the predicted Tm value of the second historical sample are matched. If the match is successful, the prediction of the second historical sample is successful; if the match fails, the prediction of the second historical sample fails, and the second historical sample that fails to match is marked as abnormal data.

[0083] The original melting curve of the second historical sample is input into the Tm value prediction model to obtain the predicted Tm value. Then, the original Tm value and the predicted Tm value are matched to measure the prediction performance of the Tm value prediction model on the test set, determining whether the model can accurately predict the Tm value corresponding to the new melting curve. If the match fails, it indicates that the model has failed to predict on that sample. This helps to identify problems and shortcomings in the Tm value prediction model, providing direction for subsequent model optimization and prompting researchers to adjust and improve the model to enhance its prediction accuracy.

[0084] By combining the previously constructed Tm value labeling model, valid samples with Tm values ​​are selected for prediction. This synergistic effect allows the subsequent prediction model to focus on processing valid data related to Tm values, reducing the interference of invalid data on the model. Simultaneously, both the labeling model and the prediction model employ the XGBoost algorithm, ensuring consistency and coherence in data processing and model construction, further improving the accuracy and reliability of the entire prediction process, resulting in more precise Tm value predictions.

[0085] One possible use case is to use 152,339 data points out of 186,138 data points as training data and 26,307 data points as testing data. Each data point contains 151 melting peak curve data points, encoded Tm values, and data source traceability information. In multi-class classification tasks, the number of encoded data points is used as the number of classifiers to train, i.e., the number of encoders is N, and the multi-class model consists of N binary classification models.

[0086] The tests revealed that setting the XGBoost model's hyperparameters as follows: a maximum of 5000 trees in the ensemble, a maximum depth of 10 per tree, a learning step size of 0.05 for weight updates in each iteration, and an objective function that optimizes probability values ​​between 0 and 1, along with using a logarithmic loss function to evaluate the model's performance, yielded good results. The test results showed that out of 27466 data points in the test set, 46 failed to correctly predict the value of Tm; the overall prediction accuracy for the test set was 99.2%.

[0087] Third embodiment:

[0088] The difference from the second embodiment is that, during the training and testing phases, the second historical sample set is composed of the first historical samples originally labeled with Tm values ​​from the first historical sample set.

[0089] In practical use, the data information of the test sample does not include the original label of whether or not it has a Tm value. The effective sample input into the Tm value prediction model is the test sample with a Tm value identified by the Tm value labeling model, so as to realize the synergy between the Tm value labeling model and the Tm value prediction model.

[0090] In a specific embodiment, the historical data information of each first historical sample in the aforementioned first historical sample set is preprocessed, including removing first historical samples without Ct values.

[0091] Samples without Tm values ​​indicate no viral infection; samples without Ct values ​​indicate an experimental failure, and these samples are of no reference value. The Tm value labeling model was cleaned.

[0092] Fourth embodiment:

[0093] Based on any of the above embodiments, the data information is preprocessed using the aforementioned Tm value labeling model. Specific steps include:

[0094] The aforementioned data also includes information on the data source and experimental process of the samples to be tested;

[0095] The above data is cleaned to remove information about the data source and experimental process of the sample to be tested;

[0096] The original dissolution curves in the above data were normalized.

[0097] While information about the data source and experimental process of the test samples is helpful in understanding the background and experimental conditions, it does not directly contribute to the determination of the Tm value and may even become a confounding factor. Removing this information during the data cleaning stage allows the model to focus more on the original dissolution curve data, which is closely related to the Tm value. Data in the original dissolution curve may have different dimensions and numerical ranges. Data with different dimensions can make it difficult for the model to balance the importance of various features during training, leading to slower convergence or even failure to converge. Normalizing the original dissolution curve, mapping all data to a specific interval (such as [0, 1] or [-1, 1]), eliminates the influence of dimensions, makes the features numerically comparable, and allows the model to adjust parameters more efficiently during training, accelerating convergence and improving training efficiency.

[0098] Fifth embodiment:

[0099] Based on any of the above embodiments, upon receiving a traceability signal, the data source information and experimental process information of the sample to be tested are obtained from the above data information.

[0100] It associates patient information, experimental parameters (well number, reagent batch, PCR instrument channel) and operation records (detection time, operator ID, original curve file) with each sample, and realizes reverse tracing of "diagnosis result → Tm value → experimental conditions" through database indexing, supports rapid problem location (such as temperature drift caused by a certain batch of reagents), and solves the deficiency of existing technology that cannot associate Tm value with original experimental parameters (well number, channel, reagent batch).

[0101] Sixth embodiment:

[0102] Based on any of the above embodiments, the preset interval temperature is 0.05℃~0.3℃, preferably 0.3℃, and the temperature range (40.0℃~85.0℃) is divided with an interval of 0.3℃ to generate a 151-bit binary vector; this improves the model's ability to capture minute temperature differences.

[0103] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the Tm value of a dissolution curve, characterized in that, Includes the following steps: Construct a Tm value labeling model and a Tm value prediction model; Obtain data information for each sample in the test sample set, and input the data information into the Tm value labeling model; wherein, the data information includes the original dissolution curve; The data information is preprocessed by the Tm value labeling model; the Tm value labeling model is used to label whether the test sample has a Tm value, and the test samples with Tm values ​​form a valid sample set; Several peak values ​​are extracted from the original dissolution curves of the test samples in the effective sample set to obtain the original Tm value set; The original Tm value set is encoded according to a preset temperature interval to obtain an original binary vector; wherein the temperature bits in the original binary vector are connected and arranged according to the temperature interval. The original melting curves of the test samples in the effective sample set are input into the Tm value prediction model to obtain a prediction binary vector; wherein, the temperature bits in the prediction binary vector are arranged in a way that is connected at intervals of temperature. The original binary vector and the predicted binary vector are compared to determine whether the predicted binary vector is abnormal.

2. The method for predicting the Tm value of the dissolution curve according to claim 1, characterized in that, The specific steps for constructing the Tm value labeling model include: Obtain historical data information of each first historical sample in the first historical sample set; wherein, the historical data information includes the original melting curve and the original label with or without Tm value label; The historical data information of each first historical sample in the first historical sample set is preprocessed; The preprocessed first historical sample set is divided into a first training set and a first test set; The historical data information from the first training set is input into the machine learning model for training to obtain the Tm value labeling model; Input the original dissolution curve of the first historical sample in the first test set into the Tm value labeling model to obtain the test label of whether the first historical sample has a Tm value label; The original label and test label of the first historical sample are matched. If the match is successful, the first historical sample is labeled normally; if the match fails, the first historical sample is labeled abnormally.

3. The method for predicting the Tm value of the dissolution curve according to claim 2, characterized in that, The specific steps for constructing a Tm value prediction model include: Obtain historical data information for each second historical sample in the second historical sample set; wherein, the historical data information includes the original dissolution curve and the original Tm value; Preprocess the historical data information of each second historical sample in the second historical sample set; The preprocessed second historical sample set is divided into a second training set and a second test set; The historical data information from the second training set is input into the machine learning model for training to obtain the Tm value prediction model. The original dissolution curve of the second historical sample in the second test set is input into the Tm value prediction model to obtain the predicted Tm value of the second historical sample. The original Tm value and the predicted Tm value of the second historical sample are matched. If the match is successful, the prediction of the second historical sample is successful; if the match fails, the prediction of the second historical sample fails.

4. The method for predicting the Tm value of the dissolution curve according to claim 3, characterized in that, The second historical sample set consists of the first historical samples in the first historical sample set that were originally labeled with Tm values.

5. The method for predicting the Tm value of the dissolution curve according to claim 3, characterized in that, The historical data information of each first historical sample in the first historical sample set is preprocessed, including removing first historical samples without Ct values.

6. The method for predicting the Tm value of the dissolution curve according to claim 1, characterized in that, The data information is preprocessed using the Tm value labeling model, and the specific steps include: The data information also includes information on the data source of the sample to be tested and information on the experimental process; The data is cleaned to remove information about the data source and experimental process of the sample to be tested. The original dissolution curves in the data are normalized.

7. The method for predicting the Tm value of the dissolution curve according to claim 6, characterized in that, Upon receiving the traceability signal, the data source information and experimental process information of the sample to be tested are obtained from the data information.

8. The method for predicting the Tm value of the dissolution curve according to claim 1, characterized in that, The preset interval temperature is 0.05℃~0.3℃.

Images